Regularization Issue with Multiple Training Datasets

[zhristophe]

I am encountering a problem with the regularization process when using two separate training datasets. Here’s the scenario:
I have two training datasets, each consisting of 10 images.
I also have a regularization dataset, which contains 20 images.

The issue arises when connecting the regularization node to the training datasets:
If I connect the regularization node to both training dataset nodes, only 10 regularization images are used, but the number of steps in one epoch is 40.
If I connect the regularization node to only one of the training dataset nodes, still only 10 regularization images are used, and the number of steps in one epoch is 30.

I would like to know if this behavior is a bug or if I am misusing the regularization feature. Thank you for your assistance.

[RaySteve312]

when you have regs, the total iterations/steps will double. even if you use only 1 reg.

[zhristophe]

I know this. The point is that only 10 reg images are used instead of 20. I have a total of 20 training images and 20 reg images, and I thought that all 20 reg images would be used.